GenZA: A General and Efficient Accelerator for Diverse Zero-Knowledge Proof Protocols

Zero-knowledge proofs (ZKPs) are widely deployed cryptographic protocols for privacy-preserving blockchains and verifiable computing. Domain-specific hardware has been proposed to accelerate its computationally expensive proof generation phase. But existing designs are limited to each single protocol and specialize in a narrow set of computational kernels, thus failing to cover the diverse algorithms that offer different desired tradeoffs for various application scenarios. Following the unified architecture design philosophy, this paper presents GenZA, a general and efficient accelerator for diverse ZKP protocols including Groth16, HyperPlonk, and Plonky2. GenZA adopts a homogeneous array of hardware units that can be flexibly reconfigured to support multiple data bitwidths from 64-bit up to 768-bit, as well as both general and special algebraic field moduli, with efficient decomposition schemes. It also incorporates a set of novel and optimized kernel mapping strategies for all the dominant kernels in the target protocols, which enable efficient execution of diverse kernels on the same hardware substrate with high resource utilization and good area efficiency. GenZA achieves two orders of magnitude speedups over CPUs, and $1.64 \times$ to $2.51 \times$ higher performance per area compared to prior ZKP accelerators, demonstrating both more general functionality support and more efficient performance across major modern ZKP protocols.